Computer vision startup Luma AI Inc. said today it has closed on a $43 million round of funding as it looks to build a new generation of generative artificial intelligence models capable of seeing the world around them.

Today’s Series B round was led by the top tier venture capital firm Andreessen Horowitz and saw participation from backers including Amplify, Matrix, Nvidia Corp. and several unnamed angel investors.

Luma began life as the creator of a simple smartphone app that eventually attracted millions of users. The app allowed people to use their smartphone camera to create 3D pictures of objects without any other hardware.

The startup is now evolving that concept, and as part of its plan it says it has created a powerful cluster of more than 3,000 of Nvidia’s A100 graphics processing units. It will be used to train new AI models that can, in the words of co-founder and Chief Technology Officer Alex Yu, “see and understand, show and explain and eventually interact with the world.”

Yu was previously an AI researcher at the University of California at Berkeley, while Luma’s other co-founder and Chief Executive Officer Amit Jain was working at Apple Inc. building out that company’s Vision Pro multimedia experiences.

Luma is looking to design and develop generative AI models that can create 3D objects from text descriptions, starting with its very first model, called Genie, which was launched in November. Going forward, the startup wants to build more sophisticated models with enhanced vision capabilities.

With Genie, users can generate interactive 3D assets using text prompts such as “Gold ring with ruby gemstone.” The model has set a high standard for fast text-to-3D generation and has already accumulated thousands of users, the startup says. Today, it’s making Genie generally available on the Luma smartphone app.

The startup is competing against a number of rivals. For instance, startups including Hypothetic, Auctoria AI Inc. and Kaedim Inc. are all building similar capabilities. The more established Stability AI Ltd. is also looking at 3D model generation, while the likes of Autodesk Inc. and Nvidia have created apps such as ClipForge, which can generate 3D models from text descriptions, and Get3D, which converts 2D images into 3D models.

Yu told TechCrunch in an interview that Luma’s 3D models will stand out with their superior fidelity. He explained that existing models for 3D image generation are trained on two-dimensional images that often “mangle spaces, bodies and movements.” As such, it can be “very difficult to generate anything coherent and usable in the first few tries.”

In a blog post, Andreessen Horowitz partners Anjney Midha and Rajko Radovanovic said Luma stands out from the crowd thanks to the abilities of its co-founders. They noted that Yu has previously authored pioneering papers on real-time neural rendering of 3D scenes and generation from a single image. Meanwhile, Jain is renowned for his work on Apple Vision Pro.

They also cited Luma’s chief scientist Jiaming Song, who is a pioneer of “diffusion models that have significantly improved state-of-the-art performance.”

Going forward, Luma will be focused on piecing together its cluster of Nvidia GPUs and simultaneously expanding its team. It said it wants to double its 24-person workforce by the end of next year.

Image: Luma AI